|
|
发表于 2025-1-22 15:28:57
|
显示全部楼层
|
The R parameter indicates the fit of the DNN model’s output to the real data, which is more than 99% according to Fig. 3(a). Choosing the first-order DNN dynamic model for the equivalent controller, unlike the feed forward ANN where the output at each moment depends on the input at the same moment, results in a smoother model output and reduces the impact of outliers and measurement noise. Additionally, the data used to estimate the DNN from the AMPC controller is collected from the real system using the KRLS one-step-forward algorithm to estimate the system model. Using the KRLS regression algorithm effectively reduces the impact of outlier data and measurement noise, resulting in cleaner data overall. |
|
[复制链接]
